Machine-learning methods for molecular states, rare events, and kinetics.

Research Statement

I develop interpretable, physics-grounded computational methods at the intersection of molecular dynamics, enhanced sampling, and machine learning. My work addresses a recurring bottleneck in molecular simulation: bridging the gap between the femtosecond timestep of atomic forces and the micro- to millisecond timescales at which biologically and chemically relevant events — conformational changes, nucleation, ligand binding — actually occur.

During my Ph.D., I built methods that accelerate this process from both ends: representation learning to identify reaction coordinates and structural phases directly from trajectory data (IceCoder), adaptive sampling to steer short parallel simulations toward rare transition pathways (PathGennie, TRAILS-MD), and kinetics estimation through weighted-ensemble resampling with temporal coherence (CoWERA).

I am motivated by problems where data-driven models must respect physical symmetries, conservation laws, and thermodynamic consistency — not just minimise loss functions. My postdoctoral research vision focuses on physics-aware molecular representation learning, scalable adaptive sampling for drug discovery, and trajectory-resolved kinetics for complex biomolecular systems.

Research Interests